Six AI workflows for business operations

Six workflows worth testing before any other

These 6 workflows repeat across fintech, healthcare and general operations teams because each one has a queue, a bottleneck and a number attached to it already. That is what makes them safe to test: you can measure the before over 5 to 10 business days, run a 14-day AI-assisted test, and measure the after on the same cases.

1. KYC document intake

KYC document intake is the step where a new customer’s identity documents, business registration and beneficial ownership information get checked against your onboarding rules before an account opens. An AI-assisted version reads the submitted documents, extracts the 10 to 15 fields your policy requires and flags anything that does not match, instead of a person retyping every field by hand.

The workflow touches personally identifying documents and, for legal entity customers, beneficial ownership data. Under the Financial Crimes Enforcement Network’s Customer Due Diligence rule, a covered financial institution must identify and verify the beneficial owners of a legal entity customer, generally anyone who owns 25% or more of the entity, before the account opens. The metric to track is time per application and the manual review rate, the share of files that still need a human to resolve.

The human checkpoint is non-negotiable here: an analyst signs off on every file the system flags, and on a random sample, commonly 1 in 10, of the other 9 files it clears automatically, to catch drift. Do not automate the actual accept or decline decision on a flagged file. Automate the document reading and field extraction that feeds that decision.

Skip this workflow if your onboarding volume is under roughly 20 applications a month. Below that, the 14-day test covers under 10 cases, not enough to read a reliable error rate, and a person reading files directly is faster than building and reviewing a test.

2. Payment dispute reconciliation

Payment dispute reconciliation matches an incoming chargeback or dispute against the original transaction, sometimes up to 90 days old, pulls the supporting evidence and drafts the response within the timeline your processor sets. Reconciliation touches transaction records, dispute correspondence and merchant evidence, all of which can include cardholder data that your PCI scope already governs.

Card network rules and, for many US consumer accounts, the Consumer Financial Protection Bureau’s Regulation E error resolution procedures set hard deadlines: a financial institution generally has 10 business days to investigate, or up to 45 days if it issues provisional credit within 10 days of the notice. An AI-assisted workflow that misses a deadline does more damage than one that runs a few hours slower, so cycle time against the regulatory clock matters as much as the error rate.

The reviewer confirms the drafted response and the matched evidence before anything posts to the case file or the cardholder. Never let the automation issue the final decision on a dispute outcome. Skip this workflow if your dispute volume is low enough that a team member already resolves each case inside the 10-day window without missing deadlines. The test only pays off once volume creates a backlog.

3. Patient intake and pre-booking triage

Patient intake and pre-booking triage sorts incoming requests, calls, forms and portal messages into 2 or 3 urgency tiers and a specialty category before a human schedules or calls the patient back. It touches protected health information from the first message, which puts it under HIPAA in the US and equivalent health-data rules elsewhere.

The metrics are time to first response and the booked share of requests, the percentage that convert to a scheduled visit without a second round of back-and-forth. Clinical staff make every decision that touches diagnosis, urgency override or care pathway. The tool sorts and drafts; it does not decide. Our companion briefing covers this workflow’s 14-day test in full, including the consent and data-handling duties involved.

Skip this workflow if your intake volume is small enough, commonly under 30 to 40 requests a day, that 1 or 2 staff already answer every message inside your target response time. Automating a queue that is not actually backed up adds risk without a matching gain.

4. Marketing reporting automation

Marketing reporting automation pulls numbers from 3 to 5 ad platforms, the CRM and an analytics tool into 1 recurring report on a fixed schedule, instead of someone assembling spreadsheets by hand every week. It touches campaign performance data and, if your CRM includes it, contact-level data that your privacy policy already covers.

The metric is hours spent on reporting per week, measured before the test starts and again after 14 days. A marketing lead approves every report before it goes to stakeholders in the first 90 days of the workflow running, then moves to spot-checking a sample once the report has held steady for 3 consecutive cycles.

Skip this workflow if your reporting already takes under 2 hours a week. Below that, the setup and review overhead can cost more than it saves, and the honest fix is trimming the report, not automating its assembly.

5. Lead qualification and routing

Lead qualification and routing scores an inbound lead against your criteria and sends it to 1 of a small number of reps or queues within 5 to 10 minutes of the form submission, instead of sitting in a shared inbox until someone checks it. It touches contact data and whatever firmographic or behavioral fields your scoring model uses.

Time to first contact and the qualified share, the percentage of routed leads a rep confirms as genuinely qualified, are the 2 numbers to track. A sales rep confirms the routing rule fired correctly on a sample of leads each week rather than reviewing every single one, since the volume here is usually too high for a full review to be practical.

Skip this workflow if 1 person already handles all inbound leads, commonly under 20 to 30 a week, and responds inside your target window. The gain from automated routing shows up once volume outpaces 1 person’s capacity to triage a shared inbox by hand.

6. Monthly close support

Monthly close support prepares reconciliations for the 5 to 10 accounts that usually cause delay, flags variances against budget and drafts the first version of close commentary, so the finance team spends its time reviewing exceptions instead of assembling the base numbers. It touches the general ledger, bank feeds and whatever subledgers feed the close.

Days to close is the metric, tracked from period end to the date the controller signs off. A close that takes 12 days can often move to 5 to 7 once the routine reconciliation work is automated and reviewed rather than performed by hand. The controller signs the close every month regardless of how the workflow performs; automation drafts the numbers, it never certifies them.

Skip this workflow if your close already finishes inside 5 days with a chart of accounts under 100 lines. The workflow pays for itself when a close currently runs past 10 days or ties up more than 1 person for most of a week.

How to sequence this if you are starting from zero

Start with the workflow where you already have a number and a complaint, not the one that sounds the most impressive in a deck. A close that takes 12 days is a louder, more measurable problem than a reporting task that takes 3 hours, so it usually deserves the first 14-day test. Run 1 workflow at a time for the first 2 rounds, roughly the first 60 to 90 days. Once a team has been through the test-review-scale cycle once, running a second workflow in parallel gets easier because the reviewer role, the rollback habit and the reporting rhythm already exist.

Every one of these 6 workflows fits under the same AI for Real Work practice: 1 named process, 1 named owner, a metric before and after. If you want the buyer-side version of this same discipline, our piece on choosing an AI automation partner covers how to scope and price an engagement around 1 of these workflows. Fintech teams weighing KYC and dispute automation against a broader operating model can read fintech marketing operations, and teams still missing an owner for their numbers should start with the marketing operations audit checklist or our piece on the revenue operations manager role.

For more on how we frame and measure this kind of work, see the AI for real work hub. If 1 of these 6 workflows matches a bottleneck you already have a number for, get in touch and we can scope a 14-day test.

FAQ

Sources

1. Financial Crimes Enforcement Network, Customer Due Diligence Rule FAQs
2. Consumer Financial Protection Bureau, Regulation E, 12 CFR 1005.11, Procedures for resolving errors
3. U.S. Department of Health and Human Services, HIPAA Minimum Necessary Requirement

Ilia PushinFounder, Pushers · Co-founder and COO, ARBI ExchangeIlia builds operating systems for growing companies in fintech and healthcare. Since 2021 he has run cross-border payments at ARBI Exchange, a licensed currency exchange in Thailand, including KYC and AML and the move into new jurisdictions.About the authorLinkedIn